Software Size Estimation Model for Board-Based Desktop Games

Abstract

Software effort, schedule and cost estimation have the highest utility at the time of inception. Since software size is one of the most important determinants of software effort (and hence cost), it is extremely beneficial to estimate size early. This early estimation of size is likely to result in better planning for projects dealing with the development of software games. Existing literature in this field contains a lot of work on software sizing for traditional software. Size estimation of software games, however, is not explored. This research addresses this gap by focusing on the size estimation of desktop based software games belonging to “board-games” category. A dataset comprising of open source board-based software desktop games is used to calibrate this size estimation model. After short listing potential predictors of software size for this sub-domain and providing definitions of them, forward step wise multiple linear regression (MLR) is used for model fitting. Results obtained show that our size estimation has reasonably better estimation accuracy as indicated by the coefficient of determination i.e. the adjusted R2 of our final model as 0.914. Mean magnitude of relative error (MMRE) is 0.24, PRED(25) is 75% and MdMRE is 0.18. Our model is validated using K-fold cross validation (K=5) and the results are very promising. Results of Kfold validation depict that the average value of PRED(25) is 72%, average MMRE is 0.23 and average MMRE is 0.18. MMER, MBRE and MIBRE are also used to assess accuracy of our model and the results obtained are up to the mark. We used object-oriented CK design metrics for the comparison of our model. Using CK metrics, another data set is formulated for the board-based software games and size estimation model is built. SLR and MLR are applied over the same data set of board games for model fitting and results are obtained. The adjusted R2 of the model obtained by using CK metrics is 0.75. Kfold validation is applied to validate the model with K=5. Comparing the results of MMRE, PRED(25), MdMRE, MMER. MBRE and MIBRE of the model obtained via CK metrics with our model show that our size estimation model has reasonable better estimation accuracy then the model built with CK metrics and hence our model can be used as a size estimation model for boardxviii based software games. The utility of this size estimation model is also demonstrated by presenting a worked-out game size estimation example followed by some size-related what-if analyses. By providing a reasonably accurate estimate of software size early in the life cycle, our model makes it easier and simpler to plan and manage the development of open source board-based software games.